A customer email does not get delivered. A prescription is written with an incorrect dosage. Two systems show different balances for the same account. Each of these errors happens because nobody verified the quality of the underlying data before using it.
Data quality dimensions are the specific, measurable characteristics of a dataset, such as accuracy, completeness, consistency, timeliness, validity, and uniqueness, that determine whether it is suitable for its intended purpose rather than merely appearing correct. Focusing on the two or three dimensions most relevant to your business goals is more valuable than trying to achieve a perfect score across all six.
That question does have an answer, and it is not a feeling. Data quality dimensions are the specific, measurable traits you examine to determine if a dataset is suitable for the task you intend it to perform. Accuracy, completeness, consistency, timeliness, validity, and uniqueness appear in almost every framework you will encounter, and together they encompass most of the practical data problems. This guide explains what each dimension means, where organisations experience the pain, how to measure them, and how to decide which should be prioritised.
Why data quality dimensions matter
You cannot improve something you do not measure, and saying "the data looks a bit off" does not constitute a measurement. The Data quality dimensions supply particular, named attributes that you can assess. These standards make certain that two individuals examining the same dataset arrive at the same conclusion instead of guessing. Lacking this shared language, quality becomes defined by the most vocal person in a meeting.
The financial impact of inaccurate data
Gartner estimates that poor data quality costs a typical organisation about $12.9 million each year. Other research shows that bad data cuts revenue markedly when the time required to correct errors is taken into account. These expenses generally do not stem from one major incident. Rather, they build up via repetitive work, for example support staff fixing the same address repeatedly. They also show up when marketing campaigns flop due to duplicate contact lists, or when finance teams spend several hours reconciling conflicting numbers that ought to match.
We have taken part in many post-mortem meetings and see the same pattern. These problems are seldom caused by a single catastrophic failure. Rather, they typically originate from a three-year-old record that was never flagged. That record sits in a join that twelve different reports now depend on.
Why this is more important for AI
A model trained on inaccurate data delivers confident but incorrect results on a large scale. Machine learning needs complete records to discover real patterns instead of statistical noise. Natural language systems require consistent formatting to pull reliable information from text fields. Recommendation engines become less effective when duplicate customer records divide one person's behaviour across several profiles. These problems do not generate error messages. Rather, they lead to AI pilots that work technically but do not get adopted because the outputs clash with the observations of experienced staff. If you are assessing whether your data is ready for an AI project, our AI and ML data readiness checklist is a useful resource to review before allocating budget.
An explanation of the six core dimensions
Most current frameworks are based on research published by Richard Wang and Diane Strong in 1996. Their work established a method for evaluating data quality based on the needs of data users rather than data storage requirements. Thirty years later, the six dimensions identified in their research remain the standard starting point for most data quality programmes.
Accuracy, completeness, and consistency
Accuracy is the degree to which a record matches reality. A customer's address is accurate if it is their actual place of residence, rather than just a correctly formatted string of text. You can verify this information by comparing it against a reliable source, such as a postal database, a bank verification process, or a confirmation from the customer.
Completeness refers to whether required fields contain values. A patient record that lacks an allergy field is incomplete for a specific reason. A doctor cannot safely prescribe medication without that information, even if all other fields are populated. Completeness depends on the requirements of the record rather than the total amount of data visible on a screen.
Consistency refers to whether a specific fact is identical in every location where it is stored. If a CRM contains one phone number for a customer while a billing system contains a different one, the data is inconsistent. This situation typically indicates that two systems update their records independently without a process to reconcile the differences.
Timeliness, validity, and uniqueness
Timeliness refers to whether data is available when it is needed and whether it reflects current conditions. A stock price from yesterday may be accurate, but it is not useful to a trader making a decision this morning. People often underrate timeliness. A dataset can pass all other quality checks but still be incorrect if it is delivered too late.
Validity is the measure of whether data follows your defined rules. Examples include checking that an email address contains an @ symbol, a postcode matches a real postal pattern, or a date falls within a logical range. These checks are typically the least expensive to automate. This makes them a practical first step for data quality, even though they cannot identify every possible error.
Uniqueness means that there are no duplicate records. Having two records for the same patient creates a risk rather than a small inconvenience. A clinician who opens one record will not see the information contained in the other. In a CRM, the consequences are less severe but the results are similar. History is split, reporting becomes confused, and a customer might receive the same marketing email twice in one morning.
These six methods provide an initial assessment of most datasets.
- Accuracy: percentage of records that match a trusted, independent source
- Completeness: percentage of required fields that hold a value
- Consistency: percentage of matching fields that agree across systems
- Timeliness: the average age of a record compared to the level of freshness your specific use case requires.
- Validity: the percentage of values that follow specific rules for format or range.
- Uniqueness: percentage of records with no duplicate elsewhere in the dataset
Extended dimensions for specific contexts
The core six cover most of what goes wrong, but not all of it, and some industries need a wider net. Integrity checks whether the relationships between pieces of data survive as information moves through your systems, so an order still points to a real customer after three system migrations rather than a customer ID that no longer resolves to anyone. Auditability matters most in regulated environments, where you need to prove that a number is right and be able to trace exactly how it got there. This overlaps heavily with data lineage, and the two are worth tackling together.
The dimension of data quality known as Relevance is frequently overlooked. It assesses whether a dataset can support a particular decision rather than merely being technically accurate. For instance, a manufacturer may discover that a supplier ID functions correctly after a system migration, yet the associated contract terms point to a version that expired two years ago. The record satisfies technical checks but is ineffective for resolving a dispute. Banks and healthcare providers also consider accessibility and security as separate quality dimensions. Data that is accurate yet unreachable by authorised users, or reachable by people who shouldn't have it, does not meet a practical requirement even when the values themselves are correct.
How to use data quality dimensions
Definitions give only limited understanding. You will probably grasp these dimensions more clearly by watching how they interact in a real setting.
Healthcare patient records
A hospital needs multiple data dimensions to keep patient safety intact during an admission. Completeness makes sure allergy fields are completed to avoid adverse reactions. Accuracy guarantees that dosages are recorded correctly, avoiding decimal errors that could render prescriptions dangerous. Uniqueness stops a patient's medical history from being divided into two records, ensuring surgeons have all necessary information before a procedure.
E-commerce customer data
Retailers rely on a particular set of requirements. Accurate delivery addresses make sure parcels reach their destination instead of being returned or refunded, which would damage customer trust. Complete contact details enable you to inform customers about delays so they are not left uninformed. Consistent data across the website, checkout system, and warehouse stops product codes from becoming mismatched. That consistency guarantees that customers receive the correct items.
Financial transaction processing
Banks give priority to timeliness and accuracy since these factors are vital to their operations. Fraud detection works only when transaction data is processed in real time. If a suspicious pattern is identified a day late, the financial loss will have already happened. An incorrect digit in an account number causes funds to be sent to the wrong recipient. This error triggers a recovery process that costs more than ensuring the data was correct from the start. Uniqueness constraints stop customers from being charged twice for a single purchase. Validity rules detect impossible transaction amounts before they impact subsequent reports.
A record needs to be available at decision time to prevent harm, even when its information is accurate and complete.
Which dimensions matter most for your use case
Treating all six dimensions as equally important results in wasted effort and limited improvements. Real-time fraud detection relies on timeliness. A perfectly accurate record arriving an hour late is useless because the action window has closed. Regulatory reporting places priority on accuracy and completeness. A gap or error in these reports leads to a compliance finding instead of a minor inconvenience. Customer segmentation mainly depends on consistency. If data is inconsistent, the same person might receive three different versions of an email because three separate systems hold different information about their identity.
Aligning your effort with business requirements rather than chasing perfect scores in every category decides whether a data quality programme yields a return or drains its budget within two years.
Data quality dimensions frequently conflict with each other in ways that frameworks do not recognise. Demanding high accuracy for a specific field usually slows data ingestion. This occurs because verifying each value against a trusted source adds a step that reduces timeliness, even when speed is the primary requirement. For example, a fraud team that validates every transaction against three external sources before flagging it places unnecessary accuracy above the speed they truly need. You should determine which dimension is your priority before adjusting the others.
Where data quality programmes go wrong
The dimensions are straightforward. Most programmes fail due to how the framework is applied rather than the framework itself.
The most common mistake is treating every dataset the same way. Measuring all six dimensions against every table in the warehouse burns weeks of engineering time on datasets nobody downstream depends on, while the one dataset that genuinely matters sits unmeasured because it wasn't first in the queue. Prioritise by business impact, not by which team shouts loudest or which dataset happens to be easiest to profile.
The second issue is attempting to achieve a perfect score. You do not need to solve the problem of ninety-nine per cent completeness on an optional field that half of your customers choose to leave blank. If you tighten a validation rule until it no longer reflects reality, you will generate false alerts. Your team will learn to ignore these notifications. When people stop responding to alerts, the monitoring system loses its value.
The third issue is not assigning ownership. A dimension score without a responsible party is merely a number on a dashboard rather than a quality programme. Every dataset you measure needs a person whose job duties include addressing the findings. That person must have the authority and time to perform the work. A simple mention in their job description is not sufficient.
The fourth mistake, and the one we see most often at Shipshape Data, is purchasing a tool before understanding the problem. Data quality platforms are effective at running checks once you know what to check and why. They are less effective at identifying which datasets matter or defining what quality levels your business requires. Teams that skip this groundwork end up with an expensive dashboard that nobody trusts.
Measuring and monitoring data quality dimensions
Scores are useful only when you can compare them over time. For this reason, you must use quantitative measurement from the start instead of relying on periodic subjective assessments.
Establish a baseline
Profile your priority datasets before you start any improvements. Record your current results, the date, and the calculation method for each score. If a customer database shows 87% accuracy, 94% completeness, and 78% uniqueness, use those figures as a baseline for your progress. You need these initial measurements to demonstrate the impact of your work over the next six months.
Automate the checks that can be automated
Manual spot-checks are not scalable and become outdated within weeks. You should establish validation rules that operate during data entry or transfer. These rules should include format checks for email and postcode fields, range checks for numerical values, and referential checks to ensure foreign keys are valid. Use fuzzy matching to identify duplicate records that contain minor differences like extra spaces or middle initials. The system must send an alert if a data quality score falls below your defined threshold. If accuracy for a critical field drops below 90%, the system should notify staff immediately so they can address the issue before customers notice it.
Place the numbers where people will notice them
A dashboard that nobody uses functions as an expensive spreadsheet. Trend lines should be shown for each dimension rather than a single snapshot. This lets you observe a gradual decline before it turns into a crisis. Separate the view by dataset or business unit so that the individuals accountable for a metric monitor its changes. The goal is for a team to spot a problem on Tuesday before a customer reports it on Friday, instead of producing an attractive chart.
Getting started
It is not necessary to measure every field in the business to derive value from this process. Select a single important dataset and evaluate it against two or three dimensions that pertain to a specific business problem. Create a functional baseline before attempting to scale the approach to additional areas. Securing success with one clean dataset is more effective for a data quality programme than building a perfect framework that remains untested.
Shipshape Data constructs the pipelines, monitoring systems, and governance structures that preserve data quality. If your team is assessing your data foundation before an AI project or a compliance deadline demands it, talk to us for assistance.
If you are presently tackling this issue on your roadmap, our data consultancy offers these services. We start by demonstrating the specific capabilities of your existing data.
Frequently asked questions
What are the data quality dimensions?
Data quality dimensions are the named, measurable attributes you examine to determine whether a dataset is suitable for its intended task, ensuring that two people reviewing the same data arrive at the same verdict rather than a shrug. The six most common are accuracy, completeness, consistency, timeliness, validity, and uniqueness, and together they encompass most of the practical data issues.
What are the six core dimensions of data quality?
The six core dimensions are accuracy, which measures how closely a record reflects reality; completeness, which monitors whether required fields contain values; consistency, which guarantees the same fact aligns across different systems; timeliness, which assesses whether data is available when needed; validity, which verifies that data adheres to defined rules; and uniqueness, which denotes the absence of duplicate records. These categories offer a standard method to evaluate most datasets.
Where do the standard data quality dimensions come from?
Most contemporary frameworks rely on research published by Richard Wang and Diane Strong in 1996. Their work introduced a method for evaluating data quality that focuses on the needs of data users rather than data storage requirements. Thirty years on, most programmes still start with the six dimensions they popularised: accuracy, completeness, consistency, timeliness, validity, and uniqueness.
What are examples of extended data quality dimensions beyond the core six?
Integrity checks that data relationships stay intact during system transfers, ensuring that an order remains linked to the correct customer after a migration. Auditability is required in regulated industries to show how specific figures were calculated, and this concept shares traits with data lineage. Relevance assesses whether a dataset is useful for a particular decision. Banks and healthcare organisations often list accessibility and security as separate dimensions of data quality as well.
Which data quality dimensions matter most for a given use case?
Real-time fraud detection relies on speed. An accurate record is useless if it arrives an hour after the event has occurred. Regulatory reporting demands accuracy and completeness because missing data leads to compliance findings. Customer segmentation needs consistency. If systems do not agree on a person's identity, that individual will receive different versions of the same email.